In [1]:
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
%matplotlib inline
In [2]:
df = pd.read_csv("Bank_Personal_Loan_Modelling.csv")
#importing dataset as "df"
df.head()
Out[2]:
In [3]:
df.info()
#exploring the datatype, columns, entries
In [4]:
print(df.shape)
#exploring the shape
In [5]:
df.describe()
#exploring the statistical summary
#Experience has a min value of "-3" (this doesn't make sense); zip code has a min of "2121.852197" (this doesn't make sense);
Out[5]:
In [6]:
df.isnull().sum()
#exploring the null values (confirms what I saw in df.info)
Out[6]:
In [7]:
df.mode().T
#The mode for "Experience" is 32 and the mode for "Zip Code" is 94720
Out[7]:
In [8]:
dupes = df.duplicated()
sum(dupes)
#no duplicates in the data
Out[8]:
In [9]:
df['ZIP Code'] = df['ZIP Code'].replace([9307], 94720)
#replacing the incorrect zip code with the mode zip code since this is more likely.
In [10]:
df.describe()
Out[10]:
In [11]:
df['Experience'] = df['Experience'].replace([-3],32)
df['Experience'] = df['Experience'].replace([-2],32)
df['Experience'] = df['Experience'].replace([-1],32)
#replaced all of the Experience years that were negative to the mode of 32 since this is more realistic.
In [12]:
df.describe()
#shows there are no invalid Zip code and not invalid experience values listed as min values.
Out[12]:
In [13]:
df.nunique()
#Number of unique in each column
Out[13]:
In [14]:
df.groupby('Mortgage').apply(lambda column: (column == 0).sum())
#shows there are 3462 people with zero mortages.
Out[14]:
In [15]:
df.groupby('CCAvg').apply(lambda column: (column == 0).sum())
#shows there are 106 people with no credit card spending per month.
Out[15]:
In [16]:
df['ZIP Code'].value_counts()
#value counts on Zip Code
Out[16]:
In [17]:
df['Education'].value_counts()
#value counts on Education
Out[17]:
In [18]:
df['Personal Loan'].value_counts()
#value counts on Personal Loan
Out[18]:
In [19]:
df['Securities Account'].value_counts()
#value counts on Securities Account
Out[19]:
In [20]:
df['CD Account'].value_counts()
#value counts on CD Account
Out[20]:
In [21]:
df['Online'].value_counts()
#value counts on Online
Out[21]:
In [22]:
df['CreditCard'].value_counts()
#value counts on CreditCard
Out[22]:
In [23]:
sns.distplot(df['Age']);
#univariate plot
In [24]:
sns.distplot(df['Experience']);
#univariate plot
In [25]:
sns.distplot(df['Income']);
#univariate plot
In [26]:
sns.distplot(df['ZIP Code']);
#univariate plot
In [27]:
sns.distplot(df['Family']);
#univariate plot
In [28]:
sns.distplot(df['CCAvg']);
#univariate plot
In [29]:
sns.distplot(df['Education']);
#univariate plot
In [30]:
sns.distplot(df['Mortgage']);
#univariate plot
In [31]:
sns.distplot(df['Personal Loan']);
#univariate plot
In [32]:
sns.distplot(df['Securities Account']);
#univariate plot
In [33]:
sns.distplot(df['CD Account']);
#univariate plot
In [34]:
sns.distplot(df['Online']);
#univariate plot
In [35]:
sns.distplot(df['CreditCard']);
#univariate plot
In [36]:
sns.jointplot(df['Income'], df['Personal Loan'])
#bivariate plot
Out[36]:
In [37]:
sns.jointplot(df['Personal Loan'], df['Age'])
#bivariate plot
Out[37]:
In [38]:
sns.pairplot(df,diag_kind='kde')
#ran pairplot
Out[38]:
In [39]:
pip install pandas-profiling
In [40]:
import pandas_profiling
pandas_profiling.ProfileReport(df)
#pandas profling to visualize the data. Reviewed Pearson's r correleation on the Correlations tab
Out[40]:
In [41]:
df2 = df.drop(['ID'], axis = 1)
#removing extra column to make it easier to work with
In [42]:
from sklearn.model_selection import train_test_split
X = df2.drop('Personal Loan',axis=1) # Predictor feature columns (8 X m)
Y = df2['Personal Loan'] # Predicted class (1=True, 0=False) (1 X m)
x_train, x_test, y_train, y_test = train_test_split(X, Y, test_size=0.3, random_state=1)
# Splitting the data into training and test sets in the ration 70:30.
x_train.head()
Out[42]:
In [43]:
print("{0:0.2f}% data is in training set".format((len(x_train)/len(df2.index)) * 100))
print("{0:0.2f}% data is in test set".format((len(x_test)/len(df2.index)) * 100))
#checking to see if the data is split
In [44]:
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
# Fit the model on train
model = LogisticRegression(solver="liblinear")
model.fit(x_train, y_train)
#predict on test
y_predict = model.predict(x_test)
#print the coefficients of all attributes
coef_df = pd.DataFrame(model.coef_)
coef_df['intercept'] = model.intercept_
print(coef_df)
In [45]:
model_score = model.score(x_test, y_test)
print(model_score)
#shows a high score
In [46]:
#Precision Score = TP / (FP + TP)
from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score
print('Precision: %.3f' % precision_score(y_test,y_predict))
In [47]:
#Recall Score = TP / (FN + TP)
print('Recall: %.3f' % recall_score(y_test, y_predict))
In [48]:
#Accuracy Score = (TP + TN) / (TP + FN + TN + FP)
print('Accuracy: %.3f' % accuracy_score(y_test, y_predict))
In [49]:
#F1 Score = 2* Precision Score * Recall Score/ (Precision Score + Recall Score/)
print('F1 Score: %.3f' % f1_score(y_test, y_predict))
In [50]:
from sklearn.metrics import roc_auc_score
y_true = np.array([0, 0, 1, 1])
y_scores = np.array([0.1, 0.4, 0.35, 0.8])
roc_auc_score(y_true, y_scores)
#ROC AUC Score
Out[50]:
In [51]:
cm=metrics.confusion_matrix(y_test, y_predict, labels=[1, 0])
df_cm = pd.DataFrame(cm, index = [i for i in ["1","0"]],
columns = [i for i in ["Predict 1","Predict 0"]])
plt.figure(figsize = (7,5))
sns.heatmap(df_cm, annot=True)
#confusion matrix showing True Positive = 44, False Postive = 33, True Negative = 1.3e+03, and False Neg = 1.0e+02.
Out[51]:
In [52]:
p=x_test.copy()
p['Observed Personal Loan Status']=y_test
p['Predicted Personal Loan Status']=y_predict
#added columns for Observed personal loan status and Predicted personal loan status
In [53]:
p.loc[p['Observed Personal Loan Status'] != p['Predicted Personal Loan Status']]
#presents the rows where the predicted personal loan status does not match the observed personal loan status
#138 rows where the predicted does not equal the observed out of 5000 rows seems like a good model.
Out[53]:
In [54]:
#Conclusion it appears that there is some coorelation between the Personal Loan and having money in a CD account. Additional correlations exist between the Personal Loan and Income and CCAvg.
#A targeted campaign for customers with a CD Account and higher income/CCAvg may be worthwhile.